Systems and methods for sample generation for identifying manufacturing defects

A system for classifying products manufactured via a manufacturing process, the system comprising:

• processor;
• and
• memory, wherein the memory has stored therein instructions that, when executed by the processor, cause the processor to:
• identify a first set of pixels associated with a first portion of a first image and a second set of pixels associated with a second portion of the first image;
• generate, via a first unsupervised deep neural network, a first mask based on the first image, wherein the first mask is associated with the first portion of the first image, wherein the generating of the first mask includes identifying a third set of pixels of the first mask that correspond to the first set of pixels of the first image;
• generate, via one or more generator neural networks for a generative adversarial network (GAN), a second image based on a noise input, wherein the second image is unlabeled;
• apply the first mask to the second image for outputting a fourth set of pixels of the second image;
• combine the fourth set of pixels of the second image with the second set of pixels that correspond to the second portion of the first image for generating a first combined image;
• predict, via a filter neural network of a defect classification system, confidence of the first combined image;
• determine that the confidence satisfies a threshold confidence;
• predict, via a defect detection neural network of the defect classification system, a classification for the first combined image;
• determine that the classification predicted for the first combined image equals an assigned classification;
• add a label to the first combined image based on the classification predicted for the first combined image;
• and
• add the first combined image with the added label to a training dataset in response to instructions that cause the processor to determine that the confidence satisfies the threshold confidence, and based on instructions that cause the processor to determine that the classification predicted for the first combined image equals the assigned classification, train a classification neural network based on the training dataset;
• apply the classification neural network to detect a defect in a manufactured product;
• and
• control a portion of a product manufacturing process based on detecting the defect.